Bulletin of the American Physical Society
APS March Meeting 2021
Volume 66, Number 1
Monday–Friday, March 15–19, 2021; Virtual; Time Zone: Central Daylight Time, USA
Session E03: Machine Learning and Data in Polymer Physics IFocus Live
|
Hide Abstracts |
Sponsoring Units: DPOLY DBIO DCOMP GSNP Chair: Debra Audus, National Institute of Standards and Technology |
Tuesday, March 16, 2021 8:00AM - 8:12AM Live |
E03.00001: Topology-Driven Completion of Chemical Data Dmitry Zubarev, Petar Ristoski Discovery of functional materials requires efficient exploration in the chemical space. We introduce an approach that identifies lacunae in the chemical data and completes them in a targeted manner. We start with topological data analysis (TDA) [1] on a set of molecules producing an approximation to Reeb graph where loops and branches are indicative of missing data. Second, we generate novel molecules that complete loops/branches on TDA graph using a modified graph-generative model for scaffold-based molecular design. The generation is conditioned on the existing scaffolds, making sure that all generated molecules contain the input scaffold. The loss function is modified to account for the generative potential of the scaffolds, gsn. We reduce the influence of the scaffolds with low gsn and penalize generation of molecules with low gsn. The application of this approach to the exploration of photo-acid generators is discussed. |
Tuesday, March 16, 2021 8:12AM - 8:24AM Live |
E03.00002: Chemically informed fragment choices to improve the property prediction for polymer systems Amanda Dumi, Daniel S. Lambrecht Disordered systems such as polymers pose particular challenges to computational modeling. The more disordered the 3-dimensional structure of a macromolecule, the more atoms must be included in a computational model to achieve representative results, which can be intractable with many accurate quantum methods. Fragmentation approaches overcome this challenge by partitioning the calculation into manageable subsystems with a subsequent merging of the results to reproduce the supersystem prediction. Fragments must be chosen judiciously to balance reduced computational scaling (smaller fragments) and accuracy (larger fragments). Fragment choice is often based on established chemical functional groups or by manual inspection, which may fail to capture important chemical interaction and can be time consuming for the user. We propose an automatic fragmentation approach in which each fragment is chosen according to quantitative criteria based on electronic-structure information achieving a systematically improvable molecular partitioning. This presentation discusses progress towards the first principles predictions of electronic polarizabilities in oligomer systems for this fragmentation approach. |
Tuesday, March 16, 2021 8:24AM - 8:36AM Live |
E03.00003: A Machine Learning Approach to the Design of Polymer Electrolyte Membranes Kuan-Hsuan Shen, Huan Tran, Chiho Kim, Rampi Ramprasad Proton exchange membrane fuel cells (PEMFCs) have been the subject of considerable research due to their potential as eco-friendly energy conversion systems. PEMFCs depend on polymer electrolyte membranes to transport protons between electrodes and thus require polymers that can readily absorb water and have sufficient proton conductivity. Among different polymeric systems, perfluorated sulfonic-acid ionomers (e.g. Nafion) are the most widely used materials, yet their high cost and low conductivity at high temperature or under low water content have led to the search for alternatives. In this work, we develop a machine learning model trained on experimental data to predict polymer water absorption and proton conductivity. We use a hierarchical fingerprinting method developed in the Polymer Genome project to represent a wide range of polymers by components over different length scales. This method not only ensures high model performance but identifies critical polymer segments or functional groups relevant to the target properties. With the model, we will screen existing and hypothetical polymers with functional groups of interest. |
Tuesday, March 16, 2021 8:36AM - 8:48AM Live |
E03.00004: Identifying Accelerated Ageing Pathways for Cross-Linked Polyethylene Pipes Through Machine Learning Joseph Damico, Melanie Hiles, Michael Grossutti, Callum Wareham, John Dutcher Cross-linked polyethylene (PEX) pipes are increasingly being used in domestic and industrial settings to transport water, gas and sewage. It is important to understand changes to the polymer and additive compounds with in-service use. We used infrared (IR) microscopy, combining the chemical specificity of IR spectroscopy with the spatial resolution of light microscopy, to track variations in the degree of crystallinity, additive concentrations as well as various chemical species across the wall thickness of PEX pipes. We have shown that principal component analysis of IR absorbance peaks can be used to classify different pipe formulations [1]. We used this methodology to characterize changes to pipes subjected to accelerated aging protocols designed to exaggerate conditions experienced by pipes during in-service use. This allowed us to identify and track IR peaks that are most relevant to pipe degradation. We used these results, together with decision tree and random forest classification algorithms, to identify different modes of pipe degradation and to better understand ageing effects on the long-term stability of PEX pipes. |
Tuesday, March 16, 2021 8:48AM - 9:00AM Live |
E03.00005: Dielectric properties of polymer nanocomposite interphases from electrostatic force microscopy using machine learning Praveen Kumar Gupta, Linda Feist Schadler, Ravishankar Sundararaman The interphase region between nanofillers and polymer matrix drastically effects the properties of nanocomposites but is hard to characterize due to nano-scale dimensions. Electrostatic force microscopy (EFM) provides a pathway to local dielectric property measurements but extracting local dielectric permittivity in complex interphase geometries from EFM measurements remains a challenge. In this work, we report a protocol of coupling experimental measurements and numerical simulations of EFM through machine learning to extract interphase dielectric permittivity in tailored silica-based nanocomposites. Silica nanoparticles were grafted with polyaniline brush of high dielectric constant to act as interphase and dispersed in polymethacrylate. We performed EFM measurements under DC polarization and generated force gradient scan across the interphase. Numerical simulations were carried out in COMSOL to match the experimental scan across this interphase to get its dielectric permittivity. Due to convolution of signals from different regions and unknown parameters in the experimental setup, we used a machine learning model to get to the best fit between the two profiles. |
Tuesday, March 16, 2021 9:00AM - 9:12AM Live |
E03.00006: Secondary structure of very large RNAs via high-throughput oligonucleotide-binding microarrays Ofer Kimchi, Rees F Garmann, Timothy Kaiwen Chiang, Megan C Engel, Vinothan N Manoharan, Michael Brenner Nucleic acid hybridization underlies an extraordinary range of in vitro biological research (e.g. CRISPR, FISH, DNA origami) as well as in vivo biological regulation (e.g. miRNA, lncRNA). In principle, hybridization is straightforward: if two nucleic acids are complementary, they will hybridize. In practice, intramolecular nucleic acid structure must be disrupted in order for hybridization to proceed, and the energetic barriers involved can preclude hybridization from occurring on realistic timescales. I will describe how we use the non-equilibrium nature of RNA hybridization as a lens to examine the equilibrium structures of large RNA molecules. We designed microarrays containing DNA oligonucleotides (oligos) perfectly complementary to different regions of structured RNA. Oligos corresponding to regions with lower secondary structure are expected to typically bind at a higher frequency, but this behavior is complicated by the fact that each hybridization involves the interaction of many nucleotides. To address this challenge, we employ high dimensional gradient descent through automatic differentiation to find the best-fit structure to the data. I will describe results of this methodology on both short RNA molecules (<100 nucleotides) as well as long RNA (>1000 nucleotides). |
Tuesday, March 16, 2021 9:12AM - 9:24AM Live |
E03.00007: Reading-out DNA translocation experiments with (un)supervised Machine Learning Ángel Díaz Carral, Magnus Ostertag, Aleksandra Radenovic, Maria Fyta DNA molecules can electrophoretically be driven through a nanoscale opening in a material giving rise to rich and measurable ionic current blockades. In this work, we train machine learning (ML) models on experimental ionic blockade data from DNA nucleotide translocation through 2D pores of different diameters. We propose a novel method that at the same time reduces the current traces to a few physical descriptors and trains low-complexity models. We describe each translocation event by four features referring to the structure of the ionic current time series. Training on this lower dimensional data and utilizing deep neural networks (DNN) and convolutional neural networks (CNN) we reach a high accuracy. Compared to more complex baseline schemes such as extreme gradient tree boosting (XGBoost) and recurrent neural network (RNN) based models trained on the full ionic current trace the former perform is comparable or even better. Our findings clearly reveal that the use of the ionic blockade height as a feature together with a proper combination of neural networks and feature extraction provides a strong enhancement in the detection and read-out sensitivity of novel nanopore sequencers. |
Tuesday, March 16, 2021 9:24AM - 9:36AM Live |
E03.00008: Understanding sequence-dependent DNA dynamics through self-associative machine learning and temperature-jump spectroscopy Mike Jones, Andrei Tokmakoff, Andrew Ferguson, Brennan Ashwood Despite rapid advances in DNA nanotechnology and a robust understanding of the associated thermodynamics, the sequence-dependent mechanisms of DNA hybridization are not fully understood. In this work, we investigate these dynamics by performing equilibrium coarse-grained simulations of oligonucleotide sequences with varied G:C placement. We employ State-Free Reversible VAMPnets to directly learn the slowest dynamical modes of each sequence and to optimize Markov State Models (MSMs) construction. Furthermore, we perform elevated temperature simulations to recapitulate temperature-jump IR and FTIR data collected on the oligonucleotides. For repetitive sequences, we find a spectrum of slow dynamics associated with out-of-register base pairing and kinetically relevant transitions between these states. In contrast, G:C pairs near the center of the duplex induce more rapid fraying dynamics. In both cases, hybridization/dissociation mechanisms deviate from an “all-or-nothing” model. Our computational predictions are in excellent accord with experiment, and provide new fundamental understanding of the sequence-dependent kinetics and mechanisms of DNA hybridization. |
Tuesday, March 16, 2021 9:36AM - 9:48AM Live |
E03.00009: Machine-guided template-based polymer retrosynthesis planning Lihua Chen, Jordan Lightstone, Rampi Ramprasad Polymer informatics is being utilized to accelerate the design and discovery of polymers. However, the practical realization of designed polymer is still slow due to synthesis challenges, e.g., difficulties with the identification of potential polymerization mechanisms and optimal reactants/solvents/processing conditions. In the past, synthesis pathways adopted for a target polymer have been heavily dependent on chemical intuition and past experience. To expedite this process, we have developed a template-based approach to assist in polymer retrosynthesis planning. In this work, several thousands of polymer polymerization paths were manually accumulated from various resources to extract hundreds of synthetic templates and used as a knowledge base. Further, a similarity metric was adopted to select the synthetic templates for the target polymer. Finally, prediction accuracy was measured by comparison with known and expert-crafted polymerization paths. We believe that the proposed template-based approach can efficiently make synthesis recommendation for experimenters, accelerating polymer discovery. |
Tuesday, March 16, 2021 9:48AM - 10:00AM Live |
E03.00010: Active Learning of Coarse Grained Models for Free Energy Surfaces Blake Duschatko, Jonathan Vandermause, Nicola Molinari, Boris Kozinsky Coarse graining procedures serve as a primary tool for alleviating limitations of all-atom molecular dynamics relating to long simulation times and large system sizes, which make simulating dynamic phenomena in polymers and proteins prohibitive. Many approaches exist for modeling the resultant interactions, but the recent progress in machine learned force fields has motivated their application to so called “bottom up” coarse grained approaches that preserve the thermodynamic properties of the all-atom system. In the present work, we explore how Gaussian processes can be used in determining the optimal complexity of coarse grained free energy surfaces. The Gaussian process framework further allows for the possibility of active learning, which has been successful in applications to ab initio molecular dynamics. We explore the extension of this framework to coarse graining problems and discuss the issue of “fine graining,” where returning to the all-atom representation is a key element of the active learning process. |
Tuesday, March 16, 2021 10:00AM - 10:12AM Live |
E03.00011: An Autonomous Liquid-Handling Platform for ML-Driven Industrial Formulation Discovery Peter Beaucage, Tyler Martin Complex liquid mixtures are the foundation of industries ranging from personal care products to biotherapeutics to specialty chemicals. While neutron and X-ray scattering methods are workhorse techniques for characterizing model formulations, the large number of components in many real products makes mapping the high-dimensional parameter space impossible due to the sheer number of possible compositions. To enable rational design of these materials, we must leverage theory, simulation, and machine learning (ML) tools to greatly reduce the expense of creating phase diagrams. Applying ML tools to scattering experiments requires a platform capable of autonomously creating and measuring samples with varying composition and chemistry. While there are numerous examples of robots which perform specific user facility operations, these systems tend to be bespoke and non-adaptable to new tasks. We have developed a highly adaptable sample environment that can be programmed to autonomously prepare and characterize liquid-formulations using neutron and X-ray scattering. Here we will highlight the design of the platform and our efforts in autonomous phase mapping of model formulations. |
Tuesday, March 16, 2021 10:12AM - 10:24AM Live |
E03.00012: Hybrid machine learning/materials science modeling for semi-crystalline polymer during film fabrication process Jian Yang, Teresa Karjala, Jonathan Mendenhall, Valeriy Ginzburg, Rajen Patel, Fawzi Hamad, Elva Lugo, Pavan Valavala For semi-crystalline polymer like polyethylene (PE), it is well known that PE film physical properties is heavily dependent on the morphology of both the crystalline phase and amorphous chains, which can be largely influence by the film processing conditions. A clear understanding of the relationships of polymer molecular fingerprint, formulation, fabrication conditions and physical properties is important for future materials design, which can be traced back to polymerization process. However, this is generally considered to be a very complicated problem due to the large parameter space. In this report, we developed a new hybrid approach to combine the power of machine learning and fundamental materials science to characterize semi-crystalline PE, develop structure-property relationship and study the effect of fabrication conditions on physical properties during blown film fabrication process and to inform the design of new polymer structures. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700